Systematizing heterogeneous expert knowledge, scenarios and goals via a goal-reasoning artificial intelligence agent for democratic urban land use planning

《Systematizing heterogeneous expert knowledge, scenarios and goals via a goal-reasoning artificial intelligence agent for democratic urban land use planning》

打印

关键字: Land use planning;Goal reasoning;Artificial intelligence;Markov decision processes;Reinforcement learning;Multicriteria decision analysis

作者单位: College of Architecture and Urban Planning, University of Tongji, Shanghai 200092, China;Department of Control Science and Engineering, Tongji University, Shanghai, 201804, China;Shanghai Institute of Intelligent Science and Technology, Tongji University, China;Harvard Graduate School of Design, Harvard University, United States;College of Architecture and Urban Planning, University of Tongji, Shanghai 200092, China;Department of Control Science and Engineering, Tongji University, Shanghai, 201804, China;Shanghai Institute of Intelligent Science and Technology, Tongji University, China;Harvard Graduate School of Design, Harvard University, United States

摘要: The tasks of democratic urban land use planning, as subjective-objective combined decision-making efforts that require considerable time and energy, have heretofore been accomplished mainly through deep human thought or by voting. In this paper, we introduce a goal-reasoning artificial intelligence (AI) agent that can assist with these tasks by combining traditional scenario planning, multicriteria decision analysis (MCDA) with a novel goal-oriented Monte Carlo tree search (G-MCTS) method. G-MCTS conducts goal-oriented searches to meet the needs of heterogeneous goals and provide the best land use solutions. We evaluated this method on a real-world planning case, and the results show that 1) the goal-reasoning AI agent is good at performing complex goal reasoning tasks with many heterogeneous expert knowledge; 2) different human planning manuscripts could be integrated into a better solution via a goal-reasoning AI agent; and 3) the goal-reasoning AI agent has the potential to make comprehensive decisions during a democratic political agenda. We conclude that the goal-reasoning AI agent, via an improved reinforcement learning (RL) method of G-MCTS, provides vast potential for assisting in subjective-objective combined urban land use planning and many other similar fields by weighing heterogeneous goals, reproducing human inspiration, and acting as a reflexive sociotechnical system.